The Agent Skills Directory

[COMMAND_EXECUTION]: Uses the herdr CLI to manage terminal panes, including listing panes, reading recent context, and running shell commands (herdr pane run). It also specifies the use of other CLI agents like cursor-agent, codex, and claude.
[EXTERNAL_DOWNLOADS]: Includes a one-time installation command npx skills add ogulcancelik/herdr which fetches code and configuration from a third-party GitHub repository not associated with a known trusted vendor.
[REMOTE_CODE_EXECUTION]: Explicitly instructs the agent to use auto-approval and permission-skipping flags such as --yolo (Cursor/Codex) and --dangerously-skip-permissions (Claude Code). This enables sub-agents to execute arbitrary commands without user intervention, increasing the potential impact of malicious input.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads output from other terminal agents to coordinate tasks. An attacker-controlled agent could output malicious instructions that the controlling agent might interpret as authoritative commands.
Ingestion points: herdr pane read <pane-id> (SKILL.md)
Boundary markers: Absent; the skill reads raw output from other panes without delimiters.
Capability inventory: Subprocess execution via herdr pane run, cursor-agent, and claude (SKILL.md).
Sanitization: Absent; the skill does not specify any validation or filtering of the ingested terminal output.

herdr